TWI286697B - Chinese opinion retrieval and extraction systems - Google Patents

Abstract

The invention concerns an opinion analysis system. It employs sentiment words to determine the opinion polarity, and detects the major topics of the related events. When users input the opinion analysis requests, the system extracts the positive and the negative opinions, and provides the summarization of the events. For the data containing temporal information, the invention presents the opinion analysis results along the time axis.

Description

1286697 IX. Description of the invention: [Technical field to which the invention pertains] The present invention relates to a technique for analyzing opinions using a computer system, in particular, using an emotional word database for opinion analysis, and providing a technical summary of the relevant event summary.

[Prior Art] ^ In an information-based society, the amount of information has increased exponentially, and the demand for users to find the required information from a large number of documents is increasing. Most of the systems currently used are search engines, user input. For information needs, search engines provide information related to information needs for users' reference. However, the traditional search engine (information retrieval system) only reflects the information related to the user's information needs, and does not further reflect that the relevant information is positive, negative, or neutral at the level of expression, and cannot be automatically analyzed. The main reason for these opinions tends.

Traditional surveys on the level of opinions are highly dependent on marketing personnel, market investigators, and public opinion, and the agencies are enchanting the specific masters. The civil and private can't decide on the subject of opinion analysis that you want to know. In the electronic file is getting more and more difficult, and the social crane towel expressed by Yi Nawei, _ arises from the computer. This can not only quickly analyze the opinions in a large amount of information, but also provide reliable opinions. Users can be free to input opinions and analysis needs. Information _ ^ Lin Information _ pure.胄讯娜纟献 can be in the electronic ^1^»* ° face information m fine, the traditional information lion% unified focus on the name of the real screaming, not listening to the level of support _ to the reference. The purpose of the analysis system is to analyze the product reputation, product reviews, online discussions, etc., and to analyze the enthusiasm of the 〇 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在= Purpose. It is hoped that the present invention can be extended to the industry ίίί and the public policy is promoted according to public opinions, and the main purpose of the present invention. I hope that this development will be related to the design, enhance people _ life and welfare, co-government, location selection products, etc. 9 128.6697 呑司汇,, & for the subject information of the relevant events.目=J invention computer system may include at least one of the following modules: , use the file to find and analyze the demand side, · = Ϊ library 'to store the file related to the theme; ΐ ΐ library' for the library with timing The document of the relationship; the database is used to store the documents after the analysis of the opinions; the database is used to store the documents with the time series relationship and after the analysis of the opinions. The library is used to store emotional words and their emotional intensity; interval; the order "engine, to classify articles according to time series, wherein the time unit can be any time-time given tif1 engine' according to the riding word' in the text Opinions on positive and negative information, and the similarity of the document nts, the vocabulary and opinions of the event, the summary and the sounds of the 総4 面 含 、 、 、 、 、 、 、 、 、 、 、 、 、 、 、 、 反^文"杈 for the sake of forbearing to describe different intensity scores, in order to respond to the support of 哎 ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ Provided for industrial use and available at Chinese brothers work well, so apply for invention patents according to law. 实施 [Embodiment] If you can let your reviewer know more about the technical content of the present invention, a preferred embodiment is described as a subject of a major implementation. Collected by the system to the relevant article 2: Ming's opinion analysis system unearths relevant positive and negative opinions, and please. Refer to the user's opinion to analyze the demand 100 as the “national card” as an example: For example, the "national card" is followed by the system file, and the file system 210 and the existing document database 220 search for the relevant cattle... and then put it into the relevant file database 400. The file I286697 stored in the related document database 400 contains the following information: file title, file text, relevant opinion analysis requirements, relevance score, and file time. Please refer to FIG. 4 for a schematic diagram of the present invention regarding the relevant number and phase. . For example: _Document-Beiku 400 cases: Document title: National card charges No consensus negotiation progress Slow document text · National card negotiation progress is slow, first card issuance fee is innocent, opinion analysis needs: national card, ..., % Aberdeen consensus... File number: sourcel_typel—19981023_0251 Relevance score: 15.950 File time: 19981023

After preparing the relevant document database 400, you can add the word breaker and the part of the word mark to enter the analysis of the body of the invention. Please refer to the schematic diagram of the analysis system first. The following is also based on the user's opinion to analyze the demand 100 as a "national card; firstly, the files in the relevant document database 400 are divided into two categories, respectively, the information group of the information and the file group without the "file time" information. : The file group of the search information is stored in the time series file database 600 according to the time series integration H desired by the user, and is stored in the time series file database 600. The file time in the relevant file is the earliest ^3 year January 1 (related The "file time" block in the file database 400 is 2〇〇3〇1〇υ, which is as late as December 31, 2003 (the file time gate ^I 20031231 in the related document database 400) as an example:弥m horse example 1 · Days as the time series unit, then divided into 365 time series file group into the time series file data Example 2: Weekly time unit, then divided into 52 time series file group into the time series file data Example 3: The month is the time series unit, and it is divided into the time series file group into the time series file database 600 0. If there is any "file time" information, you can input the opinion analysis engine 7 to detect its opinions. 700 main reference Emotional word database 11〇〇, mining related positive and negative from the file, see. “Emotional words” is an important keyword for opinion expression, including meaning, expression words, positive and negative emotional words in the emotional word database 11〇〇 And its intensity score, please refer to the u-picture is a schematic diagram of the emotional word database of the present invention. The following is an example of the content of the emotional word database 11: Example 1, positive emotional words: respectfulness front 0·3713 freq=22 ,score=〇.3426 Appreciation positive 〇·5 likes freq=6?score=〇,5〇QQ〇〇尊正freq=5,score=0.400000 重正正正freq=6,score=0.500000 许正工.反面- 0.875 Loss and face freq=37, score=-0.900000 ^ Han fteq=l8, score=-0.850000 ϋ 2 = Negative -0.51256 Dead reverse freq=64, score=-0.925120 freq=Kscore=-0.100000 1286697 Example 3, one opinion Expressing words: expressing, pointing out, proposing, emphasizing, reiterating, advocating, reflecting, declaring opinions, expressing, appealing, suggesting, thinking

In the embodiment, the scores of the emotional words are calculated by analyzing the "words" of the emotional words. Please refer to FIG. 3 for the invention of the emotional word processing bow month I engine > melon map. The basic vocabulary 105 is a commonly used basic emotional vocabulary, such as "fast j ^ machine learning, from the frequency of all positive words in the basic vocabulary 105 (hereinafter referred to as the word frequency), calculate - the emotional tendency of the word,] We use the positive tendency of the word appearing in the sub =

----Feng now the vocabulary of the positive vocabulary / positive words always appear in the frequency of the front frequency / positive words total 1 ^ ^ ^ · capsule word frequency of the opposite side of the trend = --- appear in the reverse word vocabulary frequency The reverse word always appears in the word frequency of the positive vocabulary. The word is positive. The word is the opposite of the word frequency. The total word of the word ^: the total tendency of the word = ^ (the positive tendency of the word - the opposite tendency of the word) χ confidence The index 4 Lu Xin index refers to the position of the word frequency of this word in the word frequency distribution of all words. In this embodiment, we normalize the range of word frequencies from 3 to 50 of all words to the range of 04. In the following example, the word frequency of "Zun" is 5, and the word confidence index with a word frequency of 5 falls at 0.4. That is, more than 60% of the words have a word frequency greater than 5, and 40% or less of the words have a frequency less than 5. In the end, the opinion tendencies of the vocabulary are derived from the total tendency of the words constituting the vocabulary. The following is an example of the vocabulary "Respect": The positive tendency of respect = 5/2309 = 1 〇 5/2309 + 0/5148 The opposite tendency of 尊 = 0/5148 = 〇〇 5/2309 + 0/5148 The general tendency of respect =(1-0) X 0.4 = 0.4 Heavy positive tendency = 11/2309 =0 6903 11/2309 + 11/5148 Heavy negative tendency = 11/5148 = 0.3096 11/2309 + 11/5148 * 12 1286697 Heavy The total tendency = (0·6903·0·3096) x 〇·9 = 0.3426 The tendency of the word "respect" = (the total tendency of "respect" 〇 · 4 + "heavy" total tendency 0.3426) / 2 = 0.3713 &quot ;

The opinion analysis engine 700 analyzes all the articles of each sequence unit separately (if there is no sequence unit, then all the articles are divided, and all the articles are regarded as one time series unit, and in the implementation g, there will be no sequence unit of the text early '"sequence unit" block The bit value is set to 〇), please refer to FIG. 8 for a schematic diagram of the time-series opinion file database 800 and the opinion file database 850 of the present invention. The score of each article is the sum of the positive and negative opinions in the article, and the score of each sentence is determined by the sum of the emotional word scores excavated in the sentence. The sentence opinion score calculation component 705 calculates the opinion score of each sentence, and then submits the file opinion score calculation component 710 to calculate the total opinion score of the file. Please refer to Fig. 7 for a flow chart of the opinion analysis engine 700 of the present invention. In the embodiment, the sentence is a comment sentence, and only the opinion sentence calculates the emotional word score. The town's example is represented by a box representing the opinions, and the bottom line is the emotional word: Example 1. The positive opinion of the "National Card" · 〇〇gg, the expected benefits of the promotion of the National Card project, including the New Zealand administrative, such as saving existing The cost of identity card, health insurance card new collar, reissuing, etc., the public will not need to fill in the fresh resources when you contact the public. In addition, the e-commerce application, i^ic card industry such as card reader, chip and card manufacturing Waiting for business opportunities; there are also great benefits. Z-face emotional words: promotion, efficiency, safety, promotion, and help. ---- Example 2: The negative comments of the "National Card":

00 scholars yesterday P New Zealand government fine = fine "deletion _1", _ move "job card alliance"

Negative emotional words: autocracy, totalitarianism, opposition. -- = If the sentence has a positive emotional word and a negative emotional word, then the total score after the total will determine the final opinion tendency of the sentence, such as the positive emotional word total score of 115, the negative emotional word total score of 3.8, = The total score of the sentence is 11·5_3·8=7·7, which is judged as positive; on the other hand, if the positive emotional word is 3.8, and the negative emotional word is „·5, the total score of the sentence is 3 8_115= _77, is the opposite. The opinion analysis engine 700 will completely analyze and dig out the opinions in all the articles. These thoughts, which are analyzed and analyzed, have already included the sentence layer: the owed opinion score 盥 file, The opinion scores. If there are "file time" information, they will be stored in the time series opinion file database 800, grouped according to the time series unit; if there is no "file time" information = part, it will be directly deposited into the opinion file. Database 85, please refer to Figure 8 for the present invention, see the document negative database 800 and the opinion file database 85〇. From the opinion file information f 850 and the time series opinion file database _ article, Can be clearly seen Information related to the squid sound. For example: ...Heart 13 The information file level in the chronological opinion file database 850 (Fig. 8 "Article Opinion Total Score") No.: S〇Ureeljypel —19981〇23—〇251: National Card Gray Charges No Consensus Negotiation Forced Exhibition Slow View Total Points: -10 Reverse

, 窣··意丝达辞], National Card ceremony μ〇·7〇6959) Military personal advantage (0.468159 in the body _ towel, although the money to see the traitor is D, but can also consider the text The opinion vocabulary implied by other sentences determines the tendency of the opinion. The theme detection engine 900 is based on the degree of similarity of the various opinions input, and the streamlined response causes the opinion to tend to stay at @. In the embodiment The main ^(10) file and the negative opinion file are separately summarized to show the main events that generate different opinions. Please refer to Figure 9 for the flow chart of the subject detection engine of the present invention. The subject detection engine 9〇〇:: topic vocabulary and The event vocabulary selects the scoring component 905, and selects the topic vocabulary and the event sputum in the article. The present invention selects an important event representative sentence in the document according to the topic vocabulary and the event vocabulary, wherein the main channel 5 茱 refers to being able to represent all The vocabulary set of the common theme of the document; the event vocabulary is an important sub-event that occurs under the common theme of all documents. The theory and score calculation of the topic vocabulary 舆 event vocabulary Refer to "Event Tracking based on Domaii Dependency, Fumiyo Fukumoto and Yoshimi Suzuki, SIGIR 2000 pp. 57-64". The following is an example of the topic vocabulary and event words found in the article "National Card": Example 1 , Subject vocabulary: Doubt (Na) 33.4515511550498 Agree (VK) 30.4936548784942 Item (Na) 41.7396637249344 Example 2 Event vocabulary: Slightly stipulated tactics 16.919289139043 Implementation 16.8777314125345 15.0655470437868 Research meeting 14.14486909212288 14.1082889350082 Wafer 13.81415851793844 1286697 The above is the subject vocabulary, event vocabulary and An example of a corresponding score. An article or sentence containing more of the main and event vocabulary (or a larger sum score) indicates that the article or sentence is more representative. The representative sentence selection component 910 extracts a representative sentence or article. Keke can provide relevant event information about different opinions on this topic. The following is the topic: National Card, the representative sentence information displayed by the theme detection engine 900 is an example. Extracted from the middle) K 卞 main 4 cases one, positive event generation :( sentence Figure 10 "represents the fraction" position and stopped "On behalf of the sentence" shed-bit)

10.77 10.12 9.88 Wang Lingtai: The safety of the national card is absolutely no problem. The operator: the operation will not exceed the government's entrusted scope. The national card health insurance information layer protection case is blocked.) The negative event representative sentence: (1st picture "representative score" block and "Representative sentence" 11.51 11.39 10.10 National card 6 into 4 people are afraid of venting the country, the card should be on the brink of the cliffs to review the lawful doubts of the national card policy

The abstracted event topics will be stored in the file subject database, which is a summary of the subject matter of the document. In the embodiment ϋΐί 2, after the sentence vocabulary and event vocabulary scores of the sentence are added together, after the file theme database is generated, the document file database _, the merged opinion file database Γ, r ' can generate opinion analysis information and charts It is a well-known presentation technique for the use of the knife and the chart, and will not be described here. [Effects of the Invention] The present invention can provide information related to the needs of the user's opinion analysis, in other words, (1) (2) The second party makes the opinions of the users inclined, as in the embodiment, when the opinion analysis needs "Country =" I, the system can inform the user that the general opinion tends to be positive or negative, and the month, and can also inform the user that the opinion tends to be positive or negative in a certain time interval (such as this month). Can provide users with this opinion tends to _0, such as the implementation office, when the opinion points 15 1286697 (3) (4) (5)

Mi'S,. 』_, the reason why the opinion tends to be the opposite side - is the characteristic of the National Card 6 J in the conventional technology, which is: Large:

What is the purpose of the invention and its efficacy, and not for the use of this? This "work emotion word database. Any 'change and change. The right example of the invention is repaired [simple description of the figure] The first picture is the present issue _ in opinion split, unified display · 2nd picture, this BRIEF DESCRIPTION OF THE DRAWINGS FIG. 3 is a flow chart of an emotional word processing engine of the present invention. FIG. 4 is a schematic diagram of a related document database of the present invention. FIG. 5 is a flow chart of the present invention for a timing integration engine. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 7 is a flow chart of a time series file database. FIG. 7 is a flow chart of the opinion analysis engine of the present invention. FIG. 9 is a schematic diagram of a time series opinion file database and an opinion file database. 10 is a schematic diagram of a document subject database of the present invention. FIG. 11 is a schematic diagram of an emotional word database of the present invention. [Main component symbol description] Opinion analysis requirement 100 basic vocabulary 105 demand processing engine 2 〇〇 search engine 205 Instant file database 210 existing file database 220 emotional word processing engine 300 emotional word learning component 305 emotional word weight computing component 31 〇 related documents Library 400 16 1286697 Timing Integration Engine 500 Timing Grouping Component 505 Timing File Library 600 Opinion Analysis Engine 700 Sentence Opinion Score Calculation Component 705 Document Opinion Score Calculation Element 710 Timing Opinion Document Database 8〇〇 Opinion Document Database 850 Theme Detection Engine 900

Topic vocabulary and event vocabulary selection scoring component 905 representative sentence selection component 910 file subject database 1000 emotional word database 11 〇〇 10, the scope of application for patents: • A computer system for information, including the following modules: Is the tendency to judge the emotional words and to calculate the opinions of the emotional words, 5 forms ϊ, the input of the emotional processing engine is any Chinese vocabulary, and according to the vocabulary ΐ each word appears in all known positive vocabulary And the proportion in the negative vocabulary' to judge the ίϊίίί - emotional word, - negative emotional word, or - neutral emotional word, that is, the word 之22 匕 the processing engine and calculate the _ opinion score corresponding to the word capsule, that is, the In the vocabulary, the information about the emotional tendency and emotional intensity of the Chinese vocabulary is stored in a sensation and a core. The vocabulary of the vocabulary and the emotional processing engine provide the corresponding 2·1! 2=the computer system of the description, Including: According to the emotional words, in the - file to find out the relevant 咅 π 讯 ,, and extinction - the article's opinion is divided into sentence sentence new 1 ^ 1 see the rebellious 3 · I? The computer system further includes: ff having the highest subject and opinion 3 plus =|==^ most of the subject and opinion words or one or more of the plurality of files being representative sentences to summarize the single-text 4. For example, the computer file mentioned in item 2 of the scope of patent application is the name of the spoon. The file of the opinion file is stored in the file analyzed by #Uf擎. 17

Claims (1)

1286697 Timing Integration Engine 500 Timing Grouping Component 505 Timing File Repository 600 Opinion Analysis Engine 700 Sentence Opinion Score Calculation Component 705 Document Opinion Score Calculation Element 710 Timing Opinion Document Database 8〇〇 Opinion Document Database 850 Theme Detection Engine 900 Topic vocabulary and event vocabulary selection scoring component 905 representative sentence selection component 910 file subject database 1000 emotional word database 11 〇〇 10, the scope of application for patents: • A computer system for information, including the following modules: Is the tendency to judge the emotional words and to calculate the opinions of the emotional words, 5 forms ϊ, the input of the emotional processing engine is any Chinese vocabulary, and according to the vocabulary ΐ each word appears in all known positive vocabulary And the proportion in the negative vocabulary' to judge the ίϊίίί - emotional word, - negative emotional word, or - neutral emotional word, that is, the word 之22 匕 the processing engine and calculate the _ opinion score corresponding to the word capsule, that is, the In the vocabulary, the information about the emotional tendency and emotional intensity of the Chinese vocabulary is stored in a sensation and a core. The vocabulary of the vocabulary and the emotional processing engine provide the corresponding 2·1! 2=the computer system of the description, Including: According to the emotional words, in the - file to find out the relevant 咅 π 讯 ,, and 灭 - 参 相 相 相 相 相 相 相 1 1 1 1 1 1 1 1 The computer system described in the first aspect includes: ff having the highest subject and opinion 3 plus =|==^ most of the subject and opinion words or one or more of the plurality of documents being representative sentences to summarize the bills - 4. The computer 糸# spoon surname as described in item 2 of the patent application scope. A file database of opinions, stored in the file analyzed by #Uf擎. 17 1286697 5. The computer system of claim 2, further comprising: a time-based opinion file database storing files having a timing relationship and analyzed by the opinion analysis engine. 18 1286697 VII. Designation of Representative Representatives (1) The representative representative of the case is: (1). (2) A brief description of the symbol of the representative figure: Opinion analysis requirement 100 basic vocabulary 105 demand processing engine 200 emotional word processing engine 300 related file database 400 timing integration engine 500 time series file database 600 opinion analysis engine 700 time series opinion file database 800 opinion file database 850 theme detection engine 900 Document Subject Database 1000 Emotional Word Database 1100 8. If there is a chemical formula in this case, please reveal the chemical formula Z that best shows the characteristics of the invention. 8